!pip install opendatasets --upgrade

Requirement already satisfied: opendatasets in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (0.1.20)
Requirement already satisfied: tqdm in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from opendatasets) (4.62.3)
Requirement already satisfied: click in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from opendatasets) (8.0.3)
Requirement already satisfied: kaggle in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from opendatasets) (1.5.12)
Requirement already satisfied: colorama in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from click->opendatasets) (0.4.4)
Requirement already satisfied: six>=1.10 in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from kaggle->opendatasets) (1.16.0)
Requirement already satisfied: certifi in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from kaggle->opendatasets) (2021.10.8)
Requirement already satisfied: python-dateutil in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from kaggle->opendatasets) (2.8.2)
Requirement already satisfied: requests in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from kaggle->opendatasets) (2.26.0)
Requirement already satisfied: python-slugify in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from kaggle->opendatasets) (5.0.2)
Requirement already satisfied: urllib3 in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from kaggle->opendatasets) (1.26.7)
Requirement already satisfied: text-unidecode>=1.3 in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from python-slugify->kaggle->opendatasets) (1.3)
Requirement already satisfied: idna<4,>=2.5 in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from requests->kaggle->opendatasets) (3.3)
Requirement already satisfied: charset-normalizer~=2.0.0 in c:\users\naveen\appdata\local\programs\python\python39\lib\site-packages (from requests->kaggle->opendatasets) (2.0.9)

WARNING: You are using pip version 21.2.3; however, version 21.3.1 is available.
You should consider upgrading via the 'C:\Users\Naveen\AppData\Local\Programs\Python\Python39\python.exe -m pip install --upgrade pip' command.

import opendatasets as od

Importing Libraries

%matplotlib inline
import numpy as np
import pandas as pd
import seaborn as sns
import os
import time
import random 
import shutil
import math
import torch
import pandas as pd
import torchvision
from torchvision.datasets import ImageFolder
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms, utils
from torchvision import datasets, transforms
import torch.nn.functional as F
from torchvision.utils import make_grid
import torch.nn as nn
from PIL import Image
import matplotlib.pyplot as plt
import cv2
import warnings
warnings.filterwarnings("ignore")

plt.ion()   # interactive mode

print("using Pytorch version", torch.__version__)

using Pytorch version 1.10.1+cpu

ls

 Volume in drive C is Windows
 Volume Serial Number is B8C0-2582

 Directory of C:\Users\Naveen\Downloads

17-01-2022  02:11    <DIR>          .
11-01-2022  20:48    <DIR>          ..
11-01-2022  20:26    <DIR>          .ipynb_checkpoints
13-12-2021  10:30       192,467,033 09.mkv
08-12-2021  18:09               132 2021_12_08180829output.csv
08-12-2021  22:15               121 2021_12_08221525output.csv
09-12-2021  09:14               121 2021_12_09091407output.csv
10-12-2021  19:01               121 2021_12_10190101output.csv
09-01-2022  21:40             9,785 3.jpg
22-12-2021  22:30           246,592 558515211_2021_A01 (1).pdf
23-11-2021  14:34            52,776 558515211_2021_A01.csv
23-11-2021  14:33           246,592 558515211_2021_A01.pdf
23-11-2021  14:34            42,024 558515211_2021_A01.xlsx
09-01-2022  18:48           129,434 6.jpg
10-01-2022  21:59         1,533,613 7z2107-x64.exe
03-11-2021  09:40            98,189 aaa.docx
05-12-2021  16:25         4,402,890 assgn2.zip
13-10-2021  16:12           778,932 assign1.html
13-10-2021  16:13           898,791 assign2.html
16-11-2021  14:42           444,864 assign2Markdown.pdf
08-12-2021  22:50           882,382 Assign3.html
17-12-2021  16:46             2,420 Assignment 4.Rmd
06-10-2021  12:26             5,480 Assignment1.zip
17-12-2021  16:46           779,752 Assignment-4.html
03-12-2021  19:08         1,508,304 assignment4_naveen.pdf
23-12-2021  14:51       199,015,128 AtomSetup-x64.exe
23-12-2021  14:02       551,505,186 bdline_essh_gb.zip
10-01-2022  13:39        22,779,898 Can UK host FIFA World Cup.pptx
22-12-2021  22:17           138,240 CLIENT INPUT SHEET - BASIC DETAILS & INCOME SUMMARY..xls
02-01-2022  22:37            17,235 convolutional-neural-network-pet-photo-rate.ipynb
11-01-2022  21:18         5,891,191 covid-19 vs Pneumonia vs normal analysis.html
11-01-2022  21:40         1,577,622 covid19-diagnosis-using-vgg16-resnet-and-densenet.ipynb
22-12-2021  18:05            19,782 data_ireland.kml
22-12-2021  19:24        87,092,916 Database Notes (1).pdf
15-12-2021  19:49        87,092,916 Database Notes.pdf
16-12-2021  21:12         2,208,760 data-transformation.pdf
16-12-2021  21:12           504,140 data-wrangling-cheatsheet.pdf
07-12-2021  18:25       959,896,454 downloads_backup.zip
10-01-2022  16:18        32,483,565 draft_Naveen.mp4
07-12-2021  16:44             3,583 draft1.txt
17-01-2022  02:10         2,808,344 Drafted report.docx
08-01-2022  11:29             5,792 england football.jpg
23-12-2021  13:10     1,981,473,633 england-latest-free.shp.zip
23-12-2021  12:32    <DIR>          EnglishFootballGrounds
23-12-2021  12:32             3,298 EnglishFootballGrounds.zip
31-08-2021  15:11             6,963 EPF.pdf
10-09-2021  20:44    <DIR>          FIFA 21 Installer
10-09-2021  20:38    37,487,857,948 FIFA 21 Installer.zip
05-12-2021  13:31            19,727 Figure_1.png
16-12-2021  21:09            66,909 final_2017.pdf
17-12-2021  00:39            68,970 final_2018.pdf
05-12-2021  13:23            33,098 final_model_graph.png
17-12-2021  09:01           742,083 final21.html
17-12-2021  01:26            99,993 finalsol.pdf
22-12-2021  18:20           631,799 footbal_stadiums.csv
24-11-2021  20:21            48,092 football.csv
23-11-2021  14:50           126,509 FORM 16_signed_2021-22.pdf
08-01-2022  11:14           228,763 GettyImages-453347919.jpg
01-10-2021  09:52           104,620 Graduate Programmes.xlsx
06-10-2021  12:25         1,062,279 h1 (1).pdf
12-12-2021  20:48         1,062,279 h1 (2).pdf
12-12-2021  20:48         1,062,279 h1 (3).pdf
29-09-2021  22:05         1,062,279 h1.pdf
12-12-2021  20:48            93,533 h2 (1).pdf
10-10-2021  08:03            63,136 h2.pdf
26-11-2021  13:13            57,467 h4.pdf
13-11-2021  20:55           798,782 helpful_hints.html
01-01-2022  18:16           330,400 hospital.geojson
11-01-2022  20:21    <DIR>          image dataset
11-01-2022  20:21       165,557,612 image dataset.zip
22-12-2021  13:37       514,952,619 ireland-and-northern-ireland-latest-free.shp.zip
01-11-2021  11:13        33,908,132 JAGS-4.3.0 (1).exe
16-01-2022  21:56           143,284 lecture-03-scan.pdf
16-01-2022  21:57           101,617 lecture-04-scan.pdf
16-01-2022  22:06           977,418 lecture-05-scan.pdf
16-01-2022  22:08           974,902 lecture-06-scan.pdf
16-01-2022  22:08           109,578 lecture-07-scan.pdf
16-01-2022  22:09           101,489 lecture-08-scan.pdf
16-01-2022  22:09            64,235 lecture-09-scan.pdf
16-01-2022  22:09            48,299 lecture-10-scan.pdf
16-01-2022  22:52            67,394 lecture-11-scan.pdf
16-01-2022  22:53            77,403 lecture-12-scan.pdf
16-01-2022  22:53            50,777 lecture-13-scan.pdf
16-01-2022  22:53            72,096 lecture-14-scan.pdf
16-01-2022  22:54           498,585 lecture-15-16-slides-ica.pdf
16-01-2022  22:54            35,761 lecture-15-scan.pdf
16-01-2022  22:54            20,884 lecture-16-scan-gaussian.pdf
16-01-2022  22:54            98,674 lecture-17-scan.pdf
16-01-2022  22:54            75,659 lecture-18-scan.pdf
16-01-2022  22:54           148,907 lecture-20-scan.pdf
16-01-2022  22:54            71,701 lecture-22-scan.pdf
02-12-2021  17:26         2,579,570 LinearModelAssign-4 (2).pdf
08-10-2021  22:07         1,384,687 lm.pdf
11-10-2021  15:59         1,289,615 lm_assignment1 (1).pdf
08-10-2021  22:15         1,289,615 lm_assignment1.pdf
08-10-2021  22:13         1,289,792 lm_organized.pdf
31-12-2021  22:57         2,429,480 london restaurants.geojson
31-12-2021  20:07            85,724 london_airports.geojson
01-01-2022  00:42         3,473,221 london_hotels.geojson
01-01-2022  13:26           160,858 london_hotels1.geojson
01-01-2022  14:31         1,808,213 london_hotels2.geojson
29-12-2021  11:56         1,354,754 london_stadiums.geojson
13-01-2022  00:29           590,641 main (1).html
13-01-2022  09:36             4,060 main (1).py
02-12-2021  20:02             3,070 main (1).R
13-01-2022  00:30           590,641 main (2).html
13-01-2022  00:38           590,590 main (3).html
13-01-2022  09:36           594,409 main (4).html
12-01-2022  15:16           584,512 main.html
13-01-2022  00:30            11,516 main.md
13-01-2022  09:34             4,141 main.py
19-08-2021  10:08        75,887,576 McAfee_Installer_serial_zrCtx56YdBTifr4vyYT3GQ2_key_affid_1289_akey.exe
09-01-2022  21:55            14,068 MicrosoftTeams-image.png
13-01-2022  09:36    <DIR>          ML Assignment
16-01-2022  22:58         3,762,577 ML class notes.pdf
11-01-2022  21:03        28,457,133 model_1_densenet.pt
11-01-2022  20:58        44,833,687 model_1_resnet.pt
11-01-2022  20:57        37,682,671 model_2.pt
12-09-2021  21:13             5,690 mow.png
03-01-2022  19:25           258,811 Mr. Naveen Ramprasad  Tenancy Agreement signed (1).pdf
03-01-2022  19:23           258,811 Mr. Naveen Ramprasad  Tenancy Agreement signed.pdf
03-01-2022  15:35           162,959 Mr. Naveen Ramprasad  Tenancy Agreement.pdf
25-11-2021  08:18           151,587 Naveen CV.pdf
03-01-2022  19:18           118,318 Naveen ID.jpeg
03-01-2022  19:20           231,494 Naveen ID-converted.pdf
07-10-2021  18:55         1,186,394 NAVEEN RAMPRASAD_401668_0 (1).pdf
07-10-2021  18:43         1,186,394 NAVEEN RAMPRASAD_401668_0.pdf
27-10-2021  19:09           835,388 NAVEEN RAMPRASAD_438426_0.pdf
12-11-2021  13:21           321,792 NAVEEN RAMPRASAD_453069_0.pdf
23-11-2021  23:54         3,563,879 NAVEEN RAMPRASAD_469677_0.pdf
22-12-2021  13:07           306,689 NAVEEN RAMPRASAD_536919_0.pdf
03-01-2022  19:20            90,189 Naveen rev ID.jpeg
03-11-2021  17:14           788,128 naveen_assign4.html
19-11-2021  20:10         2,498,274 naveen_assignment3.pdf
06-12-2021  11:25         2,468,472 naveen_bayesian_assign3 (1).pdf
30-11-2021  16:33         2,468,472 naveen_bayesian_assign3.pdf
25-10-2021  15:29           608,861 naveen_bayesian_assignment1_answers (1).pdf
27-10-2021  12:06           608,861 naveen_bayesian_assignment1_answers.pdf
23-11-2021  14:35            42,042 naveen_fy2020.xlsx
10-01-2022  13:59        90,640,720 OBS-Studio-27.1.3-Full-Installer-x64.exe
11-09-2021  19:02           741,546 only passport front page.pdf
09-09-2021  12:07     1,706,577,306 optiver-realized-volatility-prediction.zip
22-12-2021  22:29           224,441 OpTransactionHistoryTpr23-12-2021.pdf
10-09-2021  20:54        63,656,424 OriginThinSetup.exe
03-01-2022  11:58           989,776 osgeo4w-setup.exe
11-01-2022  21:16        23,928,832 pandoc-2.16.2-windows-x86_64.msi
11-10-2021  07:30           114,169 passport address page.pdf
11-10-2021  07:30           140,510 passport front page (1).pdf
11-09-2021  19:04           383,931 passport front page.pdf
02-01-2022  22:43     1,030,728,211 petfinder-pawpularity-score.zip
27-09-2021  20:02       147,935,384 pgadmin4-5.7-x64.exe
24-12-2021  16:17       157,942,904 pgadmin4-6.3-x64.exe
30-09-2021  06:31       128,236,408 Postman-win64-8.11.1-Setup.exe
17-10-2021  16:52           134,647 PPSN (1).PNG
09-12-2021  00:15         7,105,260 ppts.pptx
02-01-2022  22:05    <DIR>          preprocessed_data.csv
02-01-2022  22:05           495,182 preprocessed_data.csv.zip
09-09-2021  12:16       390,035,696 pycharm-community-2021.2.1.exe
09-09-2021  12:22        28,895,456 python-3.9.7-amd64.exe
04-11-2021  08:23       409,500,392 QGIS-OSGeo4W-3.16.12-1-Setup-x86_64.exe
03-01-2022  12:34     1,062,685,696 QGIS-OSGeo4W-3.16.15-1.msi
08-12-2021  22:50             2,344 ques3.txt
10-12-2021  16:06             4,098 r.R
17-12-2021  02:31        14,138,526 R_All_slides.pdf
17-12-2021  02:31            30,154 r_exam_doc.xlsx
09-09-2021  21:14        89,550,304 R-4.1.1-win.exe
23-12-2021  12:49           104,288 RAMPRASAD NAVEEN_BGNPN4873J_FY201920_16_SIGNED.pdf
16-01-2022  19:29         2,597,100 Report (1).docx
03-01-2022  08:32         5,746,082 restaurants.geojson
13-09-2021  21:40            48,068 Resume-2018.docx
09-09-2021  21:18       156,183,344 RStudio-1.4.1717.exe
23-12-2021  13:02       377,245,127 scotland-latest-free.shp.zip
10-01-2022  14:43            21,431 script.docx
05-12-2021  10:04            87,776 sonar.all-data
07-12-2021  18:25    <DIR>          srikanthan_tirumala-hooper-s3-proxy-9de6a38098ef
07-12-2021  18:23           148,049 srikanthan_tirumala-hooper-s3-proxy-9de6a38098ef.zip
30-11-2021  16:48         1,134,357 ST405_Assignment3.html
30-11-2021  16:44             7,457 ST405_Assignment3.Rmd
09-12-2021  13:50    <DIR>          ST661 exam 2020, with solution-20211209
09-12-2021  13:49           252,532 ST661 exam 2020, with solution-20211209.zip
15-12-2021  09:55         1,173,121 St661 Group Project Presentation.pptx
10-12-2021  16:28         1,540,224 ST661_group7_report.html
10-12-2021  16:27            29,980 ST661_group7_report.Rmd
24-12-2021  21:26             9,931 statistic_id865968_population-of-scotland-2020-by-council-area.xlsx
24-12-2021  20:04            10,857 statistic_id971694_population-of-england-in-2020-by-county.xlsx
05-10-2021  21:50           964,624 stats_assignment1_answers (1).pdf
07-10-2021  18:43           964,624 stats_assignment1_answers (2).pdf
10-10-2021  11:00           964,624 stats_assignment1_answers (3).pdf
05-10-2021  15:48           904,168 stats_assignment1_answers.pdf
12-09-2021  21:49            17,314 Student-CV-Template.docx
03-12-2021  18:56           731,543 test.html
10-12-2021  01:36             3,422 test.Rmd
07-12-2021  18:27                12 test_output.csv.csv
07-01-2022  20:38            46,278 test1.sql
10-12-2021  10:36            21,759 tf_draft (2).Rmd
10-12-2021  11:42            21,773 tf_draft (3).rmd
10-12-2021  10:55            23,481 tf_draft _js1
10-12-2021  10:33         1,068,795 tf_draft.html
10-12-2021  10:36            18,891 tf_draft.Rmd
10-12-2021  10:33    <DIR>          tf_draft_files
10-12-2021  10:38            23,481 tf_draft_js
09-12-2021  16:20             5,302 tf_draft_mg
10-12-2021  11:48            28,748 tf_draft_mg1
10-12-2021  13:13            31,219 tf_draft_mg2
10-12-2021  11:47            28,748 tf_draft_naveens_version (1).Rmd
10-12-2021  16:03            29,979 tf_draft_naveens_version.Rmd
08-01-2022  14:13           114,088 the-four-countries-of-the-united-kingdom.jpg
24-11-2021  16:13            21,387 tut.php
24-12-2021  20:01         2,576,896 ukpopestimatesmid2020on2021geography.xls
04-11-2021  08:21         5,095,424 Unconfirmed 707591.crdownload
11-10-2021  07:30           186,996 visa.jpeg
16-01-2022  23:12        42,762,464 vlc-3.0.16-win64.exe
03-01-2022  18:37           348,992 WebLaunchRecorder.exe
27-09-2021  17:31            44,721 WhatsApp Image 2021-09-27 at 6.31.02 PM.jpeg
05-10-2021  20:23           153,214 WhatsApp Image 2021-10-05 at 9.22.39 PM.jpeg
05-10-2021  20:24           177,217 WhatsApp Image 2021-10-05 at 9.22.39 PM-converted.pdf
05-10-2021  20:24         1,081,170 WhatsApp Image 2021-10-05 at 9.22.39 PM-converted-merged.pdf
25-10-2021  15:25           111,072 WhatsApp Image 2021-10-25 at 4.25.32 PM.jpeg
25-10-2021  15:27           111,931 WhatsApp Image 2021-10-25 at 4.25.32 PM-converted.pdf
25-10-2021  15:27           678,118 WhatsApp Image 2021-10-25 at 4.25.32 PM-converted-merged.pdf
19-11-2021  19:57            61,259 WhatsApp Image 2021-11-19 at 7.56.22 PM (1).jpeg
19-11-2021  19:57            71,832 WhatsApp Image 2021-11-19 at 7.56.22 PM (2).jpeg
19-11-2021  19:57            78,075 WhatsApp Image 2021-11-19 at 7.56.22 PM (3).jpeg
19-11-2021  19:57            67,249 WhatsApp Image 2021-11-19 at 7.56.22 PM (4).jpeg
19-11-2021  19:57            77,123 WhatsApp Image 2021-11-19 at 7.56.22 PM (5).jpeg
19-11-2021  19:58            73,890 WhatsApp Image 2021-11-19 at 7.56.22 PM (6).jpeg
19-11-2021  19:58            57,255 WhatsApp Image 2021-11-19 at 7.56.22 PM (7).jpeg
19-11-2021  19:58            68,234 WhatsApp Image 2021-11-19 at 7.56.22 PM (8).jpeg
19-11-2021  19:57            69,064 WhatsApp Image 2021-11-19 at 7.56.22 PM.jpeg
10-01-2022  21:57         1,016,632 winzip26-downwz.exe
29-11-2021  00:38         1,016,632 winzip26-home.exe
             220 File(s) 48,544,929,052 bytes
              11 Dir(s)  106,506,862,592 bytes free

Preparing Training and Test Sets

class_names = ['Normal', 'Viral Pneumonia', 'Covid']
data_dir = 'image dataset/Covid19-dataset'

Creating the data loaders

The dataset is a set of x-RAY images of normal patients,patients having covid and patients having pneumonia.Hence this is a 3 class classification problem

train_transform = transforms.Compose([
    transforms.Resize(size=(224, 224)),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])
test_transform = transforms.Compose([
    transforms.Resize(size=(224, 224)),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])

train_ds = ImageFolder(data_dir+'/train', train_transform)
test_ds = ImageFolder(data_dir+'/test', test_transform)
classes = train_ds.classes

classes

['Covid', 'Normal', 'Viral Pneumonia']

def describe_dataset(data):
    classes_count = { c:0 for c in classes}
    for _,label in data:
        classes_count[classes[label]]+=1

    print(f'Classes: {len(classes)} - {classes} ')
    print(f'Examples: {len(data)}')
    print(f'Counts: {classes_count}')
    plt.bar(classes_count.keys(),classes_count.values(), alpha=0.3 )

    return classes_count

_ = describe_dataset(train_ds)

Classes: 3 - ['Covid', 'Normal', 'Viral Pneumonia'] 
Examples: 251
Counts: {'Covid': 111, 'Normal': 70, 'Viral Pneumonia': 70}

from torch.utils.data import random_split
random_seed = 42
torch.manual_seed(random_seed);

val_size = 30
train_size = len(train_ds) - val_size

train_ds, val_ds = random_split(train_ds, [train_size, val_size])

print("\nTRAIN DATASET")
_=describe_dataset(train_ds)

print("\nVALIDATION DATASET")
_=describe_dataset(val_ds)

print("\nTEST DATASET")
_=describe_dataset(test_ds)

dataset = None
new_dataset = None

TRAIN DATASET
Classes: 3 - ['Covid', 'Normal', 'Viral Pneumonia'] 
Examples: 221
Counts: {'Covid': 97, 'Normal': 65, 'Viral Pneumonia': 59}

VALIDATION DATASET
Classes: 3 - ['Covid', 'Normal', 'Viral Pneumonia'] 
Examples: 30
Counts: {'Covid': 14, 'Normal': 5, 'Viral Pneumonia': 11}

TEST DATASET
Classes: 3 - ['Covid', 'Normal', 'Viral Pneumonia'] 
Examples: 66
Counts: {'Covid': 26, 'Normal': 20, 'Viral Pneumonia': 20}

batch_size = 5

train_loader = torch.utils.data.DataLoader(train_ds, batch_size=batch_size, shuffle=True, num_workers=3, pin_memory=True)
test_loader = torch.utils.data.DataLoader(test_ds, batch_size=batch_size, shuffle=True)
valid_loader = torch.utils.data.DataLoader(val_ds, batch_size=batch_size, shuffle=True)

print('Number of training batches', len(train_loader))
print('Number of validation batches', len(valid_loader))
print('Number of test batches', len(test_loader))

Number of training batches 45
Number of validation batches 6
Number of test batches 14

Data Visualization

def show_images(images, labels, preds):
    plt.figure(figsize=(8, 4))
    for i, image in enumerate(images):
        plt.subplot(1, 6, i + 1, xticks=[], yticks=[])
        image = image.numpy().transpose((1, 2, 0))
        mean = np.array([0.485, 0.456, 0.406])
        std = np.array([0.229, 0.224, 0.225])
        image = image * std + mean
        image = np.clip(image, 0., 1.)
        plt.imshow(image)
        col = 'green'
        if preds[i] != labels[i]:
            col = 'red'
            
        plt.xlabel(f'{classes[int(labels[i].numpy())]}')
        plt.ylabel(f'{classes[int(preds[i].numpy())]}', color=col)
        if(i==5):
            break
    plt.tight_layout()
    plt.show()

images, labels = next(iter(train_loader))

images.shape

torch.Size([5, 3, 224, 224])

show_images(images, labels, labels) #passing the predicitons as the labels

print(train_ds[0][0].shape)
train_ds[0]

torch.Size([3, 224, 224])

(tensor([[[-1.8268, -1.8097, -1.8097,  ..., -1.7069, -1.6898, -1.7069],
          [-1.8439, -1.8268, -1.8268,  ..., -1.6898, -1.6898, -1.7240],
          [-1.8439, -1.8439, -1.8268,  ..., -1.6898, -1.6898, -1.7069],
          ...,
          [-1.7583, -1.7240, -1.7240,  ..., -1.8782, -1.8610, -1.8610],
          [-1.7754, -1.7240, -1.7240,  ..., -1.8610, -1.8782, -1.8782],
          [-1.7412, -1.7240, -1.7240,  ..., -1.8610, -1.8782, -1.8782]],
 
         [[-1.7381, -1.7206, -1.7206,  ..., -1.6155, -1.5980, -1.6155],
          [-1.7556, -1.7381, -1.7381,  ..., -1.5980, -1.5980, -1.6331],
          [-1.7556, -1.7556, -1.7381,  ..., -1.5980, -1.5980, -1.6155],
          ...,
          [-1.6681, -1.6331, -1.6331,  ..., -1.7906, -1.7731, -1.7731],
          [-1.6856, -1.6331, -1.6331,  ..., -1.7731, -1.7906, -1.7906],
          [-1.6506, -1.6331, -1.6331,  ..., -1.7731, -1.7906, -1.7906]],
 
         [[-1.5081, -1.4907, -1.4907,  ..., -1.3861, -1.3687, -1.3861],
          [-1.5256, -1.5081, -1.5081,  ..., -1.3687, -1.3687, -1.4036],
          [-1.5256, -1.5256, -1.5081,  ..., -1.3687, -1.3687, -1.3861],
          ...,
          [-1.4384, -1.4036, -1.4036,  ..., -1.5604, -1.5430, -1.5430],
          [-1.4559, -1.4036, -1.4036,  ..., -1.5430, -1.5604, -1.5604],
          [-1.4210, -1.4036, -1.4036,  ..., -1.5430, -1.5604, -1.5604]]]),
 1)

def show_example(img, label):
    print('Label: ', classes[label], "("+str(label)+")")
    plt.imshow(img.permute(1, 2, 0))

show_example(*train_ds[0])

Label:  Normal (1)

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

show_example(*train_ds[90])

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

Label:  Viral Pneumonia (2)

def show_batch(dl):
    for images, labels in dl:
        fig, ax = plt.subplots(figsize=(24, 12))
        ax.set_xticks([]); ax.set_yticks([])
        ax.imshow(make_grid(images, nrow=16).permute(1, 2, 0))
        break

show_batch(train_loader)

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

show_batch(test_loader)

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

show_batch(valid_loader)

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

Training and Evaluation functions

def change_device(input):
    if torch.cuda.is_available():
        return input.cuda()
    else:
        return input

def train(model, train_loader):
    
    model.train()
    
    epoch_loss = 0

    for train_step, (images, labels) in enumerate(train_loader):

        x_train = change_device(images)
        y_train = change_device(labels)

        optimizer.zero_grad()
        
        y_pred = model(x_train)
        
        loss = loss_fn(y_pred,y_train)
        
        loss.backward()
        
        optimizer.step()

        if(train_step%20==0):
            print(f"batch:   {train_step}   train_loss   {loss.item()}   perplexity   = {torch.exp(loss)}")
        
        epoch_loss += loss.item()
        
    return epoch_loss / (train_step +1 )

def evaluate(model, loader):
    
    model.eval()
    print("validation/test set evaluation")
    epoch_loss = 0
    
    with torch.no_grad():
        accuracy = 0
        for val_step, (images, labels) in enumerate(loader):
            x_train = change_device(images)
            y_train = change_device(labels)

            y_pred = model(x_train)

            loss = loss_fn(y_pred,y_train)

            _, preds = torch.max(y_pred, 1)
            accuracy += sum(( preds.cpu() == labels).numpy())

            if(val_step%3==0):
                print(f"batch:   {val_step}   validation_loss   {loss.item()}   perplexity   = {torch.exp(loss)}")
        
            epoch_loss += loss.item()
        accuracy = accuracy/len(val_ds)
        print(f' Accuracy: {accuracy:.4f}')

    return epoch_loss / (val_step + 1)

def epoch_time(start_time, end_time):
    elapsed_time = end_time - start_time
    elapsed_mins = int(elapsed_time / 60)
    elapsed_secs = int(elapsed_time - (elapsed_mins * 60))
    return elapsed_mins, elapsed_secs

def show_preds(model,loader):
    model.eval() 
    print("test set evaluation")

    with torch.no_grad():
        accuracy = 0
        for test_step, (images, labels) in enumerate(loader):
            x_train = change_device(images)
            y_train = change_device(labels)

            y_pred = model(x_train)

            loss = loss_fn(y_pred,y_train)

            _, preds = torch.max(y_pred, 1)
          
            accuracy += sum(( preds.cpu() == labels).numpy())
        accuracy = accuracy/len(test_ds)
        print(f' Accuracy: {accuracy:.4f}')

    show_images(images, labels, preds.cpu())

Simple implementation of VGG

class VGGNet(nn.Module):
    def __init__ (self):
        super(VGGNet,self).__init__()

        self.conv1 = nn.Conv2d(in_channels=3,out_channels=64,kernel_size=3,padding=1,stride=1) #224
        self.maxPool1 = nn.MaxPool2d(2,2)

        self.conv2 = nn.Conv2d(in_channels=64,out_channels=128,kernel_size=3,padding=1,stride=1) #112
        self.maxPool2 = nn.MaxPool2d(2,2)

        self.conv3 = nn.Conv2d(in_channels=128,out_channels=256,kernel_size=3,padding=1,stride=1) #56
        self.maxPool3 = nn.MaxPool2d(2,2)

        self.conv4 = nn.Conv2d(in_channels=256,out_channels=512,kernel_size=3,padding=1,stride=1) #28
        self.maxPool4 = nn.MaxPool2d(2,2)

        self.conv5 = nn.Conv2d(in_channels=512,out_channels=512,kernel_size=1,padding=0,stride=1)
        self.maxPool5 = nn.MaxPool2d(2,2)

        self.conv6 = nn.Conv2d(in_channels=512,out_channels=512,kernel_size=3,padding=1,stride=1) #14
        self.maxPool6 = nn.MaxPool2d(2,2)

        self.fc1 = nn.Linear(4608,1024) #7
        self.fc2 = nn.Linear(1024,512)
        self.fc3 = nn.Linear(512,3)

    def forward(self , x):
               
        x = self.maxPool1(F.relu(self.conv1(x)))

        x = self.maxPool2(F.relu(self.conv2(x)))

        x = self.maxPool3(F.relu(self.conv3(x)))

        x = self.maxPool4(F.relu(self.conv4(x)))

        x = self.maxPool5(F.relu(self.conv5(x)))

        x = self.maxPool6(F.relu(self.conv6(x)))

        x = x.reshape(x.shape[0], -1)

        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = F.relu(self.fc3(x))

        return x

model = change_device(VGGNet())

model

VGGNet(
  (conv1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (maxPool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv2): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (maxPool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv3): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (maxPool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv4): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (maxPool4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv5): Conv2d(512, 512, kernel_size=(1, 1), stride=(1, 1))
  (maxPool5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv6): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (maxPool6): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (fc1): Linear(in_features=4608, out_features=1024, bias=True)
  (fc2): Linear(in_features=1024, out_features=512, bias=True)
  (fc3): Linear(in_features=512, out_features=3, bias=True)
)

for param in model.parameters():
    print(f' {param.size()} parameters are {param.numel()}')

 torch.Size([64, 3, 3, 3]) parameters are 1728
 torch.Size([64]) parameters are 64
 torch.Size([128, 64, 3, 3]) parameters are 73728
 torch.Size([128]) parameters are 128
 torch.Size([256, 128, 3, 3]) parameters are 294912
 torch.Size([256]) parameters are 256
 torch.Size([512, 256, 3, 3]) parameters are 1179648
 torch.Size([512]) parameters are 512
 torch.Size([512, 512, 1, 1]) parameters are 262144
 torch.Size([512]) parameters are 512
 torch.Size([512, 512, 3, 3]) parameters are 2359296
 torch.Size([512]) parameters are 512
 torch.Size([1024, 4608]) parameters are 4718592
 torch.Size([1024]) parameters are 1024
 torch.Size([512, 1024]) parameters are 524288
 torch.Size([512]) parameters are 512
 torch.Size([3, 512]) parameters are 1536
 torch.Size([3]) parameters are 3

def count_parameters(model):
    return sum(p.numel() for p in model.parameters() if p.requires_grad)

print(f'The model has {count_parameters(model):,} trainable parameters and total memory usage is {count_parameters(model)*(4/(2**20)*3)} mb')

The model has 9,419,395 trainable parameters and total memory usage is 107.79642105102539 mb

Training and testing

Learning Rate = 3e-5; Epoch = 1

loss_fn = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=3e-5)

N_EPOCHS = 1

best_valid_loss = float('inf')

for epoch in range(N_EPOCHS):
    
    start_time = time.time()
    
    train_loss = train(model, train_loader)
    valid_loss = evaluate(model, valid_loader)
    
    end_time = time.time()
    
    epoch_mins, epoch_secs = epoch_time(start_time, end_time)
    
    if valid_loss < best_valid_loss:
        print("saving the model")
        best_valid_loss = valid_loss
        torch.save(model.state_dict(), 'model_2.pt')
    
    print(f'Epoch: {epoch+1:02} | Time: {epoch_mins}m {epoch_secs}s')
    print(f'\tTrain Loss: {train_loss:.3f} | Train PPL: {math.exp(train_loss):7.3f}')
    print(f'\t Val. Loss: {valid_loss:.3f} |  Val. PPL: {math.exp(valid_loss):7.3f}')

batch:   0   train_loss   1.0941715240478516   perplexity   = 2.9867072105407715
batch:   20   train_loss   1.0990883111953735   perplexity   = 3.0014283657073975
batch:   40   train_loss   1.1033796072006226   perplexity   = 3.014336109161377
validation/test set evaluation
batch:   0   validation_loss   1.0952149629592896   perplexity   = 2.9898252487182617
batch:   3   validation_loss   1.0792638063430786   perplexity   = 2.9425125122070312
 Accuracy: 0.7333
saving the model
Epoch: 01 | Time: 1m 39s
	Train Loss: 1.098 | Train PPL:   2.997
	 Val. Loss: 1.088 |  Val. PPL:   2.968

model.load_state_dict(torch.load('model_2.pt'))

<All keys matched successfully>

show_preds(model, test_loader) #learning rate 3e-5

test set evaluation
 Accuracy: 0.5152

Hence testing out the various learning rates I felt that a learning rate of 3e-5 showed significant improvement hence decided to use that for the rest of the models

Simple Implementation of Resnet Neural Network

class two_layer_Block(nn.Module):
    def __init__(self, num_layers, in_channels,intermediate_channels, identity_downsample=None, stride=1):
        super(two_layer_Block, self).__init__()

        self.num_layers = num_layers
        self.expansion = 1

        self.conv1 = nn.Conv2d(in_channels,intermediate_channels,kernel_size=3, padding = 1, stride=1)

        self.bn1 = nn.BatchNorm2d(intermediate_channels)

        self.conv2 = nn.Conv2d(intermediate_channels,intermediate_channels,kernel_size=3, padding=1, stride= stride)

        self.bn2 = nn.BatchNorm2d(intermediate_channels)

        self.relu = nn.ReLU()

        self.identity_downsample = identity_downsample
      
    def forward(self, x):
        identity = x.clone()
        x = self.relu(self.bn1(self.conv1(x)))
        x = self.bn2(self.conv2(x))

        if self.identity_downsample is not None:
            identity = self.identity_downsample(identity)

        x += identity
        x = self.relu(x)
        return x

class three_layer_Block(nn.Module):
    def __init__(self, num_layers, in_channels,intermediate_channels, identity_downsample=None, stride=1):
        super(three_layer_Block, self).__init__()
        self.num_layers = num_layers
        self.expansion = 4

        self.conv1 = nn.Conv2d(in_channels, intermediate_channels, kernel_size=1, stride=1, padding=0)

        self.bn1 = nn.BatchNorm2d(intermediate_channels)

        self.conv2 = nn.Conv2d(intermediate_channels,intermediate_channels,kernel_size=3,stride=stride,padding=1)

        self.bn2 = nn.BatchNorm2d(intermediate_channels)

        self.conv3 = nn.Conv2d(intermediate_channels,intermediate_channels * self.expansion,kernel_size=1,stride=1,padding=0)

        self.bn3 = nn.BatchNorm2d(intermediate_channels * self.expansion)

        self.relu = nn.ReLU()

        self.identity_downsample = identity_downsample

        self.stride = stride
      
    def forward(self, x):

        identity = x.clone()
        x = self.relu(self.bn1(self.conv1(x)))
        x = self.relu(self.bn2(self.conv2(x)))
        x = self.bn3(self.conv3(x))

        if self.identity_downsample is not None:
            identity = self.identity_downsample(identity)

        x += identity
        x = self.relu(x)
        return x

class ResNet(nn.Module):
    def __init__(self,  num_layers, Block, img_channels, num_classes):
        super(ResNet, self).__init__()
        self.expansion = 4

        if ( num_layers< 50):
            self.expansion = 1

        if (num_layers==18):
            layers = [2,2,2,2]
      
        elif (num_layers==34):
            layers = [3,4,6,3]
        
        elif (num_layers == 50):
            layers = [3,4,6,3]
        elif (num_layers==101):
            layers = [3,4,23,3]
        elif (num_layers==152):
            layers = [3,8,36,3]

        print(layers)
        self.in_channels = 64
        self.conv1 = nn.Conv2d(in_channels=img_channels, out_channels=64, kernel_size=7, padding=3, stride=2)
        self.bn = nn.BatchNorm2d(64)
        self.relu = nn.ReLU()
        self.maxpool = nn.MaxPool2d(kernel_size=3,stride=2,padding=1)

        self.layer1 = self.make_layer(num_layers, Block, layers[0],intermediate_channels = 64,stride=1)
        self.layer2 = self.make_layer(num_layers, Block, layers[1],intermediate_channels = 128,stride=2)
        self.layer3 = self.make_layer(num_layers, Block, layers[2],intermediate_channels = 256,stride=2)
        self.layer4 = self.make_layer(num_layers, Block, layers[3],intermediate_channels = 512,stride=2)
        self.avgpool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Linear(512*self.expansion, num_classes)

    def forward(self, x):

        x = self.relu(self.bn(self.conv1(x)))
        x = self.maxpool(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.avgpool(x)
        x = x.reshape(x.shape[0], -1)
        x = self.fc(x)
        return x

    def make_layer(self, num_layers, block, num_residual_blocks, intermediate_channels, stride):
        layers = []
        identity_downsample = None
        identity_downsample = nn.Sequential(nn.Conv2d(self.in_channels, intermediate_channels*self.expansion, kernel_size=1, stride=stride),
                                                      nn.BatchNorm2d(intermediate_channels*self.expansion))


        layers.append(block(num_layers, self.in_channels, intermediate_channels, identity_downsample, stride))
        self.in_channels = intermediate_channels * self.expansion 

        for i in range(num_residual_blocks - 1):
            layers.append(block(num_layers, self.in_channels, intermediate_channels)) 

        return nn.Sequential(*layers)

def ResNet18(img_channels=3, num_classes=1000):
    return ResNet(18,two_layer_Block, img_channels, num_classes)


def ResNet34(img_channels=3, num_classes=1000):
    return ResNet(34, two_layer_Block, img_channels, num_classes)


def ResNet50(img_channels=3, num_classes=1000):
    return ResNet(50, three_layer_Block, img_channels, num_classes)


def ResNet101(img_channels=3, num_classes=1000):
    return ResNet(101,  three_layer_Block, img_channels, num_classes)


def ResNet152(img_channels=3, num_classes=1000):
    return ResNet(152,  three_layer_Block, img_channels, num_classes)

model = change_device(ResNet18(3,3))

[2, 2, 2, 2]

print(model)

ResNet(
  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3))
  (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (relu): ReLU()
  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
  (layer1): Sequential(
    (0): two_layer_Block(
      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
      (identity_downsample): Sequential(
        (0): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1))
        (1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): two_layer_Block(
      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
    )
  )
  (layer2): Sequential(
    (0): two_layer_Block(
      (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
      (identity_downsample): Sequential(
        (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2))
        (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): two_layer_Block(
      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
    )
  )
  (layer3): Sequential(
    (0): two_layer_Block(
      (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
      (identity_downsample): Sequential(
        (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2))
        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): two_layer_Block(
      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
    )
  )
  (layer4): Sequential(
    (0): two_layer_Block(
      (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1))
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
      (identity_downsample): Sequential(
        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2))
        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      )
    )
    (1): two_layer_Block(
      (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu): ReLU()
    )
  )
  (avgpool): AdaptiveAvgPool2d(output_size=1)
  (fc): Linear(in_features=512, out_features=3, bias=True)
)

for param in model.parameters():
    print(f' {param.size()} parameters are {param.numel()}')

 torch.Size([64, 3, 7, 7]) parameters are 9408
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64, 64, 3, 3]) parameters are 36864
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64, 64, 3, 3]) parameters are 36864
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64, 64, 1, 1]) parameters are 4096
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64, 64, 3, 3]) parameters are 36864
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64, 64, 3, 3]) parameters are 36864
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([128, 64, 3, 3]) parameters are 73728
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128, 128, 3, 3]) parameters are 147456
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128, 64, 1, 1]) parameters are 8192
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128, 128, 3, 3]) parameters are 147456
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128, 128, 3, 3]) parameters are 147456
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([256, 128, 3, 3]) parameters are 294912
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256, 256, 3, 3]) parameters are 589824
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256, 128, 1, 1]) parameters are 32768
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256, 256, 3, 3]) parameters are 589824
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256, 256, 3, 3]) parameters are 589824
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([512, 256, 3, 3]) parameters are 1179648
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512, 512, 3, 3]) parameters are 2359296
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512, 256, 1, 1]) parameters are 131072
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512, 512, 3, 3]) parameters are 2359296
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512, 512, 3, 3]) parameters are 2359296
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([3, 512]) parameters are 1536
 torch.Size([3]) parameters are 3

def count_parameters(model):
    return sum(p.numel() for p in model.parameters() if p.requires_grad)

print(f'The model has {count_parameters(model):,} trainable parameters and total memory usage is {count_parameters(model)*(4/(2**20)*3)} mb')

The model has 11,187,139 trainable parameters and total memory usage is 128.0266456604004 mb

loss_fn = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=3e-5)

N_EPOCHS = 1

best_valid_loss = float('inf')

for epoch in range(N_EPOCHS):
    
    start_time = time.time()
    
    train_loss = train(model, train_loader)
    valid_loss = evaluate(model, valid_loader)
    
    end_time = time.time()
    
    epoch_mins, epoch_secs = epoch_time(start_time, end_time)
    
    if valid_loss < best_valid_loss:
        print("saving the model")
        best_valid_loss = valid_loss
        torch.save(model.state_dict(), 'model_1_resnet.pt')
    
    print(f'Epoch: {epoch+1:02} | Time: {epoch_mins}m {epoch_secs}s')
    print(f'\tTrain Loss: {train_loss:.3f} | Train PPL: {math.exp(train_loss):7.3f}')
    print(f'\t Val. Loss: {valid_loss:.3f} |  Val. PPL: {math.exp(valid_loss):7.3f}')

batch:   0   train_loss   1.0592479705810547   perplexity   = 2.8842012882232666
batch:   20   train_loss   0.6090688109397888   perplexity   = 1.8387184143066406
batch:   40   train_loss   0.25497061014175415   perplexity   = 1.2904237508773804
validation/test set evaluation
batch:   0   validation_loss   0.2943599224090576   perplexity   = 1.3422669172286987
batch:   3   validation_loss   0.2874559462070465   perplexity   = 1.3330318927764893
 Accuracy: 0.8667
saving the model
Epoch: 01 | Time: 1m 29s
	Train Loss: 0.670 | Train PPL:   1.955
	 Val. Loss: 0.306 |  Val. PPL:   1.358

model.load_state_dict(torch.load('model_1_resnet.pt'))

<All keys matched successfully>

show_preds(model, test_loader) # using learning rate 3e-5

test set evaluation
 Accuracy: 0.7424

Simple Implementation of Densenet Neural Network

class DenseNet(nn.Module):
    def __init__(self, img_channels, num_classes, num_init_features, growth_factor,layers):
        super(DenseNet,self).__init__()

        self.growth_factor = growth_factor

        self.conv1 = nn.Conv2d(in_channels=img_channels,out_channels=num_init_features,kernel_size=7,padding=3,stride=2)

        self.bn = nn.BatchNorm2d(num_init_features)

        self.relu = nn.ReLU()

        self.maxpool = nn.MaxPool2d(kernel_size=3,stride=2,padding=1)

        self.denselayer1 = self.make_layer(DenseBlock,num_init_features, layers[0])

        self.transitionlayer1 = TransitionBlock(num_init_features + self.growth_factor*layers[0])

        self.denselayer2 = self.make_layer(DenseBlock,int((num_init_features + growth_factor*layers[0])/2), layers[1])

        self.transitionlayer2 = TransitionBlock(int((num_init_features + self.growth_factor*layers[0])/2 + self.growth_factor*layers[1]))

        self.denselayer3 = self.make_layer(DenseBlock,int(((num_init_features + self.growth_factor*layers[0])/2 + self.growth_factor*layers[1])/2), layers[2])

        self.transitionlayer3 = TransitionBlock(int(((num_init_features + self.growth_factor*layers[0])/2 + self.growth_factor*layers[1])/2) + self.growth_factor*layers[2])

        self.denselayer4 = self.make_layer(DenseBlock,int((((num_init_features + self.growth_factor*layers[0])/2 + self.growth_factor*layers[1])/2 + self.growth_factor*layers[2])/2), 
                                      layers[3])

        self.averagepool = nn.AdaptiveAvgPool2d(1)

        self.fc = nn.Linear(int(num_init_features/8 + self.growth_factor*(layers[0]/8 + layers[1]/4 + layers[2]/2 + layers[3])),num_classes) #we only need to chage the output features while pre training

    def forward(self, x):
        x = self.relu(self.bn(self.conv1(x)))
        x = self.maxpool(x)
        x = self.denselayer1(x)
        x = self.transitionlayer1(x)
        x = self.denselayer2(x)
        x = self.transitionlayer2(x)
        x = self.denselayer3(x)
        x = self.transitionlayer3(x)
        x = self.denselayer4(x)
        x = self.averagepool(x)
        x = x.reshape(x.shape[0], -1)
        x = self.fc(x)

        return x

    def make_layer(self, DenseBlock, features, layer):
        dense_block = []
        for i in range(layer):
            dense_block.append(DenseBlock(features + i*32,self.growth_factor))

        return nn.Sequential(*dense_block)

class DenseBlock(nn.Module):
    def __init__ (self,input_features, growth_factor):
        super(DenseBlock, self).__init__()
        self.bn1 = nn.BatchNorm2d(input_features)
        self.relu1 = nn.ReLU()
        self.conv1 = nn.Conv2d(in_channels=input_features,out_channels=4*growth_factor,kernel_size=1,stride=1)
        self.bn2 = nn.BatchNorm2d(4*growth_factor)
        self.relu2 = nn.ReLU()
        self.conv2 = nn.Conv2d(in_channels=4*growth_factor,out_channels=growth_factor,kernel_size=3,stride=1,padding=1)

    def forward(self, x):
        concat_layer = x.clone()
        x = self.bn1(x)
        x = self.relu1(x)
        x = self.conv1(x)
        x = self.bn2(x)
        x = self.relu2(x)
        x = self.conv2(x)

        return torch.cat((concat_layer, x), dim=1)



class TransitionBlock(nn.Module):
    def __init__(self, input_features):
        super(TransitionBlock,self).__init__()
        self.bn = nn.BatchNorm2d(input_features)
        self.relu = nn.ReLU()
        self.conv1 = nn.Conv2d(in_channels=input_features,out_channels=int(input_features/2),kernel_size=1,stride=1,padding=0)
        self.avpool = nn.AvgPool2d(kernel_size=2, stride=2)

    def forward(self, x):
        x = self.bn(x)
        x = self.relu(x)
        x = self.conv1(x)
        x = self.avpool(x)

        return x




def DenseNet121(img_channels=3, num_classes=1000,num_init_features=64,growth_factor=32):
    return DenseNet(img_channels,num_classes, num_init_features, growth_factor,[6, 12, 24, 16])

def DenseNet161(img_channels=3, num_classes=1000,num_init_features=96,growth_factor=48):
    return DenseNet(img_channels,num_classes, num_init_features, growth_factor,[6, 12, 36, 24])

def DenseNet169(img_channels=3, num_classes=1000,num_init_features=64,growth_factor=32):
    return DenseNet(img_channels,num_classes, num_init_features, growth_factor,[6, 12, 32, 32])

def DenseNet201(img_channels=3, num_classes=1000,num_init_features=64,growth_factor=32):
    return DenseNet(img_channels,num_classes, num_init_features, growth_factor,[6, 12, 48, 32])

def DenseNet264(img_channels=3, num_classes=1000,num_init_features=64,growth_factor=32):
    return DenseNet(img_channels,num_classes, num_init_features, growth_factor,[6, 12, 64, 48])

model = change_device(DenseNet121(3,3)) 

model

DenseNet(
  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3))
  (bn): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (relu): ReLU()
  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
  (denselayer1): Sequential(
    (0): DenseBlock(
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(64, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (1): DenseBlock(
      (bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(96, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (2): DenseBlock(
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (3): DenseBlock(
      (bn1): BatchNorm2d(160, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(160, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (4): DenseBlock(
      (bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(192, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (5): DenseBlock(
      (bn1): BatchNorm2d(224, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(224, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
  )
  (transitionlayer1): TransitionBlock(
    (bn): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu): ReLU()
    (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
    (avpool): AvgPool2d(kernel_size=2, stride=2, padding=0)
  )
  (denselayer2): Sequential(
    (0): DenseBlock(
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(128, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (1): DenseBlock(
      (bn1): BatchNorm2d(160, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(160, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (2): DenseBlock(
      (bn1): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(192, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (3): DenseBlock(
      (bn1): BatchNorm2d(224, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(224, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (4): DenseBlock(
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (5): DenseBlock(
      (bn1): BatchNorm2d(288, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(288, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (6): DenseBlock(
      (bn1): BatchNorm2d(320, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(320, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (7): DenseBlock(
      (bn1): BatchNorm2d(352, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(352, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (8): DenseBlock(
      (bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(384, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (9): DenseBlock(
      (bn1): BatchNorm2d(416, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(416, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (10): DenseBlock(
      (bn1): BatchNorm2d(448, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(448, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (11): DenseBlock(
      (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(480, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
  )
  (transitionlayer2): TransitionBlock(
    (bn): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu): ReLU()
    (conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1))
    (avpool): AvgPool2d(kernel_size=2, stride=2, padding=0)
  )
  (denselayer3): Sequential(
    (0): DenseBlock(
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (1): DenseBlock(
      (bn1): BatchNorm2d(288, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(288, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (2): DenseBlock(
      (bn1): BatchNorm2d(320, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(320, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (3): DenseBlock(
      (bn1): BatchNorm2d(352, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(352, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (4): DenseBlock(
      (bn1): BatchNorm2d(384, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(384, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (5): DenseBlock(
      (bn1): BatchNorm2d(416, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(416, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (6): DenseBlock(
      (bn1): BatchNorm2d(448, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(448, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (7): DenseBlock(
      (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(480, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (8): DenseBlock(
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (9): DenseBlock(
      (bn1): BatchNorm2d(544, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(544, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (10): DenseBlock(
      (bn1): BatchNorm2d(576, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(576, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (11): DenseBlock(
      (bn1): BatchNorm2d(608, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(608, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (12): DenseBlock(
      (bn1): BatchNorm2d(640, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(640, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (13): DenseBlock(
      (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(672, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (14): DenseBlock(
      (bn1): BatchNorm2d(704, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(704, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (15): DenseBlock(
      (bn1): BatchNorm2d(736, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(736, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (16): DenseBlock(
      (bn1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(768, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (17): DenseBlock(
      (bn1): BatchNorm2d(800, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(800, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (18): DenseBlock(
      (bn1): BatchNorm2d(832, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(832, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (19): DenseBlock(
      (bn1): BatchNorm2d(864, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(864, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (20): DenseBlock(
      (bn1): BatchNorm2d(896, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(896, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (21): DenseBlock(
      (bn1): BatchNorm2d(928, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(928, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (22): DenseBlock(
      (bn1): BatchNorm2d(960, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(960, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (23): DenseBlock(
      (bn1): BatchNorm2d(992, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(992, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
  )
  (transitionlayer3): TransitionBlock(
    (bn): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (relu): ReLU()
    (conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1))
    (avpool): AvgPool2d(kernel_size=2, stride=2, padding=0)
  )
  (denselayer4): Sequential(
    (0): DenseBlock(
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (1): DenseBlock(
      (bn1): BatchNorm2d(544, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(544, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (2): DenseBlock(
      (bn1): BatchNorm2d(576, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(576, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (3): DenseBlock(
      (bn1): BatchNorm2d(608, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(608, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (4): DenseBlock(
      (bn1): BatchNorm2d(640, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(640, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (5): DenseBlock(
      (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(672, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (6): DenseBlock(
      (bn1): BatchNorm2d(704, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(704, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (7): DenseBlock(
      (bn1): BatchNorm2d(736, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(736, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (8): DenseBlock(
      (bn1): BatchNorm2d(768, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(768, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (9): DenseBlock(
      (bn1): BatchNorm2d(800, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(800, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (10): DenseBlock(
      (bn1): BatchNorm2d(832, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(832, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (11): DenseBlock(
      (bn1): BatchNorm2d(864, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(864, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (12): DenseBlock(
      (bn1): BatchNorm2d(896, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(896, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (13): DenseBlock(
      (bn1): BatchNorm2d(928, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(928, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (14): DenseBlock(
      (bn1): BatchNorm2d(960, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(960, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
    (15): DenseBlock(
      (bn1): BatchNorm2d(992, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu1): ReLU()
      (conv1): Conv2d(992, 128, kernel_size=(1, 1), stride=(1, 1))
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
      (relu2): ReLU()
      (conv2): Conv2d(128, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
    )
  )
  (averagepool): AdaptiveAvgPool2d(output_size=1)
  (fc): Linear(in_features=1024, out_features=3, bias=True)
)

for param in model.parameters():
    print(f' {param.size()} parameters are {param.numel()}')

 torch.Size([64, 3, 7, 7]) parameters are 9408
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([64]) parameters are 64
 torch.Size([128, 64, 1, 1]) parameters are 8192
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([96]) parameters are 96
 torch.Size([96]) parameters are 96
 torch.Size([128, 96, 1, 1]) parameters are 12288
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128, 128, 1, 1]) parameters are 16384
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([160]) parameters are 160
 torch.Size([160]) parameters are 160
 torch.Size([128, 160, 1, 1]) parameters are 20480
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([192]) parameters are 192
 torch.Size([192]) parameters are 192
 torch.Size([128, 192, 1, 1]) parameters are 24576
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([224]) parameters are 224
 torch.Size([224]) parameters are 224
 torch.Size([128, 224, 1, 1]) parameters are 28672
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([128, 256, 1, 1]) parameters are 32768
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128, 128, 1, 1]) parameters are 16384
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([160]) parameters are 160
 torch.Size([160]) parameters are 160
 torch.Size([128, 160, 1, 1]) parameters are 20480
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([192]) parameters are 192
 torch.Size([192]) parameters are 192
 torch.Size([128, 192, 1, 1]) parameters are 24576
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([224]) parameters are 224
 torch.Size([224]) parameters are 224
 torch.Size([128, 224, 1, 1]) parameters are 28672
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([128, 256, 1, 1]) parameters are 32768
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([288]) parameters are 288
 torch.Size([288]) parameters are 288
 torch.Size([128, 288, 1, 1]) parameters are 36864
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([320]) parameters are 320
 torch.Size([320]) parameters are 320
 torch.Size([128, 320, 1, 1]) parameters are 40960
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([352]) parameters are 352
 torch.Size([352]) parameters are 352
 torch.Size([128, 352, 1, 1]) parameters are 45056
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([384]) parameters are 384
 torch.Size([384]) parameters are 384
 torch.Size([128, 384, 1, 1]) parameters are 49152
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([416]) parameters are 416
 torch.Size([416]) parameters are 416
 torch.Size([128, 416, 1, 1]) parameters are 53248
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([448]) parameters are 448
 torch.Size([448]) parameters are 448
 torch.Size([128, 448, 1, 1]) parameters are 57344
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([480]) parameters are 480
 torch.Size([480]) parameters are 480
 torch.Size([128, 480, 1, 1]) parameters are 61440
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([256, 512, 1, 1]) parameters are 131072
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([256]) parameters are 256
 torch.Size([128, 256, 1, 1]) parameters are 32768
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([288]) parameters are 288
 torch.Size([288]) parameters are 288
 torch.Size([128, 288, 1, 1]) parameters are 36864
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([320]) parameters are 320
 torch.Size([320]) parameters are 320
 torch.Size([128, 320, 1, 1]) parameters are 40960
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([352]) parameters are 352
 torch.Size([352]) parameters are 352
 torch.Size([128, 352, 1, 1]) parameters are 45056
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([384]) parameters are 384
 torch.Size([384]) parameters are 384
 torch.Size([128, 384, 1, 1]) parameters are 49152
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([416]) parameters are 416
 torch.Size([416]) parameters are 416
 torch.Size([128, 416, 1, 1]) parameters are 53248
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([448]) parameters are 448
 torch.Size([448]) parameters are 448
 torch.Size([128, 448, 1, 1]) parameters are 57344
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([480]) parameters are 480
 torch.Size([480]) parameters are 480
 torch.Size([128, 480, 1, 1]) parameters are 61440
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([128, 512, 1, 1]) parameters are 65536
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([544]) parameters are 544
 torch.Size([544]) parameters are 544
 torch.Size([128, 544, 1, 1]) parameters are 69632
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([576]) parameters are 576
 torch.Size([576]) parameters are 576
 torch.Size([128, 576, 1, 1]) parameters are 73728
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([608]) parameters are 608
 torch.Size([608]) parameters are 608
 torch.Size([128, 608, 1, 1]) parameters are 77824
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([640]) parameters are 640
 torch.Size([640]) parameters are 640
 torch.Size([128, 640, 1, 1]) parameters are 81920
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([672]) parameters are 672
 torch.Size([672]) parameters are 672
 torch.Size([128, 672, 1, 1]) parameters are 86016
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([704]) parameters are 704
 torch.Size([704]) parameters are 704
 torch.Size([128, 704, 1, 1]) parameters are 90112
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([736]) parameters are 736
 torch.Size([736]) parameters are 736
 torch.Size([128, 736, 1, 1]) parameters are 94208
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([768]) parameters are 768
 torch.Size([768]) parameters are 768
 torch.Size([128, 768, 1, 1]) parameters are 98304
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([800]) parameters are 800
 torch.Size([800]) parameters are 800
 torch.Size([128, 800, 1, 1]) parameters are 102400
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([832]) parameters are 832
 torch.Size([832]) parameters are 832
 torch.Size([128, 832, 1, 1]) parameters are 106496
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([864]) parameters are 864
 torch.Size([864]) parameters are 864
 torch.Size([128, 864, 1, 1]) parameters are 110592
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([896]) parameters are 896
 torch.Size([896]) parameters are 896
 torch.Size([128, 896, 1, 1]) parameters are 114688
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([928]) parameters are 928
 torch.Size([928]) parameters are 928
 torch.Size([128, 928, 1, 1]) parameters are 118784
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([960]) parameters are 960
 torch.Size([960]) parameters are 960
 torch.Size([128, 960, 1, 1]) parameters are 122880
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([992]) parameters are 992
 torch.Size([992]) parameters are 992
 torch.Size([128, 992, 1, 1]) parameters are 126976
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([1024]) parameters are 1024
 torch.Size([1024]) parameters are 1024
 torch.Size([512, 1024, 1, 1]) parameters are 524288
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([512]) parameters are 512
 torch.Size([128, 512, 1, 1]) parameters are 65536
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([544]) parameters are 544
 torch.Size([544]) parameters are 544
 torch.Size([128, 544, 1, 1]) parameters are 69632
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([576]) parameters are 576
 torch.Size([576]) parameters are 576
 torch.Size([128, 576, 1, 1]) parameters are 73728
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([608]) parameters are 608
 torch.Size([608]) parameters are 608
 torch.Size([128, 608, 1, 1]) parameters are 77824
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([640]) parameters are 640
 torch.Size([640]) parameters are 640
 torch.Size([128, 640, 1, 1]) parameters are 81920
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([672]) parameters are 672
 torch.Size([672]) parameters are 672
 torch.Size([128, 672, 1, 1]) parameters are 86016
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([704]) parameters are 704
 torch.Size([704]) parameters are 704
 torch.Size([128, 704, 1, 1]) parameters are 90112
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([736]) parameters are 736
 torch.Size([736]) parameters are 736
 torch.Size([128, 736, 1, 1]) parameters are 94208
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([768]) parameters are 768
 torch.Size([768]) parameters are 768
 torch.Size([128, 768, 1, 1]) parameters are 98304
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([800]) parameters are 800
 torch.Size([800]) parameters are 800
 torch.Size([128, 800, 1, 1]) parameters are 102400
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([832]) parameters are 832
 torch.Size([832]) parameters are 832
 torch.Size([128, 832, 1, 1]) parameters are 106496
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([864]) parameters are 864
 torch.Size([864]) parameters are 864
 torch.Size([128, 864, 1, 1]) parameters are 110592
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([896]) parameters are 896
 torch.Size([896]) parameters are 896
 torch.Size([128, 896, 1, 1]) parameters are 114688
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([928]) parameters are 928
 torch.Size([928]) parameters are 928
 torch.Size([128, 928, 1, 1]) parameters are 118784
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([960]) parameters are 960
 torch.Size([960]) parameters are 960
 torch.Size([128, 960, 1, 1]) parameters are 122880
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([992]) parameters are 992
 torch.Size([992]) parameters are 992
 torch.Size([128, 992, 1, 1]) parameters are 126976
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([128]) parameters are 128
 torch.Size([32, 128, 3, 3]) parameters are 36864
 torch.Size([32]) parameters are 32
 torch.Size([3, 1024]) parameters are 3072
 torch.Size([3]) parameters are 3

def count_parameters(model):
    return sum(p.numel() for p in model.parameters() if p.requires_grad)

print(f'The model has {count_parameters(model):,} trainable parameters and total memory usage is {count_parameters(model)*(4/(2**20)*3)} mb')

The model has 6,965,123 trainable parameters and total memory usage is 79.70950698852539 mb

loss_fn = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=3e-5)

N_EPOCHS = 1

best_valid_loss = float('inf')

for epoch in range(N_EPOCHS):
    
    start_time = time.time()
    
    train_loss = train(model, train_loader)
    valid_loss = evaluate(model, valid_loader)
    
    end_time = time.time()
    
    epoch_mins, epoch_secs = epoch_time(start_time, end_time)
    
    if valid_loss < best_valid_loss:
        print("saving the model")
        best_valid_loss = valid_loss
        torch.save(model.state_dict(), 'model_1_densenet.pt')
    
    print(f'Epoch: {epoch+1:02} | Time: {epoch_mins}m {epoch_secs}s')
    print(f'\tTrain Loss: {train_loss:.3f} | Train PPL: {math.exp(train_loss):7.3f}')
    print(f'\t Val. Loss: {valid_loss:.3f} |  Val. PPL: {math.exp(valid_loss):7.3f}')

batch:   0   train_loss   1.0867397785186768   perplexity   = 2.9645931720733643
batch:   20   train_loss   0.9385002851486206   perplexity   = 2.556144952774048
batch:   40   train_loss   0.9804096221923828   perplexity   = 2.6655478477478027
validation/test set evaluation
batch:   0   validation_loss   0.5349668264389038   perplexity   = 1.707391619682312
batch:   3   validation_loss   0.859047532081604   perplexity   = 2.3609108924865723
 Accuracy: 0.5667
saving the model
Epoch: 01 | Time: 3m 38s
	Train Loss: 0.900 | Train PPL:   2.459
	 Val. Loss: 0.630 |  Val. PPL:   1.878

show_preds(model, test_loader)

test set evaluation
 Accuracy: 0.5909

Summary and Model Results evaluation

Since denset pre trained obtained the higest results we will use that for further analysis

Data Table for the training data </H4> The test and the validation dataset have 30 data each

import plotly.figure_factory as ff

data_matrix = [['Classes','Data_available'],
               ['Normal', 70],
               ['Viral_Pneumonia', 70],
               ['Covid', 111]]

colorscale = [[0, '#4d004c'],[.5, '#f2e5ff'],[1, '#ffffff']]

fig = ff.create_table(data_matrix,colorscale=colorscale)
fig.update_layout(
    title_text = 'Data used for training',
    margin = {'t':40},
    height = 200,
    title_x=0.5
)
fig.show()

Accuracy Table

import plotly.figure_factory as ff

data_matrix = [['Testing paramter','Self-VGG-16', 'Self-Resnet','Self-Densenet'],
               ['Accuracy', 0.5152, 0.7424, 0.5909]]

fig = ff.create_table(data_matrix)
fig.show()